Forthcoming in Erkenntnis Meaning and Formal Semantics in Generative Grammar Stephen Schiffer New York University The history of semantics in generative linguistics raises questions. By the time Chomsky’s Aspects of Syntactic Theory was published in 1965 generative grammars were understood to have a semantic component in addition to a syntactic and phonological component, and it was assumed that a speaker’s knowledge of her language required her to have tacit knowledge of a generative grammar of it. At that time the question that defined semantics in linguistics was the form that the internally represented semantic theory should take, and that is what the defining question was taken to be in Katz and Fodor’s seminal 1963 manifesto, “The Structure of a Linguistic Theory,” the first serious effort to do semantics in generative linguistics. Then around 1970 linguistic semantics took a curious turn. Without rejecting the claim that speaking a language requires tacit knowledge of a semantic theory of it, linguists turned away from the project of characterizing the nature of that theory in order to pursue instead the Montague-inspired project of providing for the languages we speak the same kind of formal semantics that logicians devise for the artificial languages of formal systems of logic. ‘Formal semantics’ originally signified semantics for formal languages devised for the mathematical study of formal systems of logic, but the expression now has a meaning akin to ‘analytical philosophy’ and signifies the Montague-inspired approach to the semantical study of natural languages. At the same time, many theorists—including many formal semanticists—recognize that the theories semanticists construct under the formal semantics rubric can’t plausibly be regarded as theories of the kind needed to explain a speaker’s knowledge of her language. The obvious question this bifurcation raises concerns the relation between, on the one hand, the psychologically explanatory semantic theories still thought to be needed but no longer the object of study in linguistic semantics and, on the other hand, the theories formal semanticists are concerned to construct. That question, I shall argue, becomes urgent when we understand the way in which a psychologically explanatory theory would supersede even the best formal- semantic theory. *** I. The Generative Grammar Hypothesis A speaker of a natural language has the ability to understand indefinitely many sentences of her language that she has never previously encountered; indeed, her ability to understand any sentence of her language does not depend on her having a prior acquaintance with it. What explains this remarkable ability? One familiar answer emanating from MIT would require what I shall call the Generative Grammar Hypothesis (GGH): The ability of a speaker of a natural language L to understand sentences of L requires her to have tacit knowledge of a generative grammar of L, that being a finitely specifiable theory of L that generates one or more syntactic structures for each sentence of L and interprets those structures both phonologically and semantically.1 Of course, while the sentence one understands may be novel, neither the words that compose it nor the syntactical rules by which it’s composed are novel, and it’s commonly thought to be a platitude that one knows what a sentence means by way of knowing what its words and syntactic structure mean. If, as GGH tells us, one tacitly knows a generative grammar of one’s language, then that would throw considerable light on how the platitude comes to be true. No doubt GGH derives much of its appeal from its ability to explain the platitude. We can’t hope to assess GGH other than relative to some way of understanding the connected notions of sentence understanding and semantic interpretation as they occur in GGH. GGH’s relation to the (apparent) platitude about meaning suggests that 1 In saying that GGH “emanates from MIT” I don’t intend to be representing any particular linguist’s views, let alone some time slice of Noam Chomsky. Rather, I take GGH to encapsulate a view that has a certain prima facie appeal and is close enough to what many theorists accept to sustain the conclusions I shall reach in this essay. The notion of a generative grammar utilized in GGH is, however, the view Chomsky accepted in his (1965). 2 understanding a sentence is knowing what it means, and that, consequently, it must be the job of the semantic component of a generative grammar to issue, for each sentence of the language, in a theorem that tells us what the sentence means. That is indeed true, as far as it goes, but it’s helpful only to the extent that we already know in what knowing a sentence’s meaning consists; it doesn’t relieve us of having to ask what way of understanding sentence understanding and semantic interpretation has the best chance of verifying GGH. Before turning to that and related questions, however, I should say a little something now about the parts of GGH that for present purposes I will assume not to be contentious. Linguists like to speak of a language user’s “internal grammar”; but a grammar is a theory, and theories are abstract things that can’t be in anyone’s head or anywhere else. Better to speak, as linguists also do, of a person’s internally represented grammar. The notion of tacit knowledge of a grammar is intended to make talk of an internally represented grammar somewhat more precise. One has tacit knowledge of a proposition p when one is in a subpersonal, or subdoxastic,2 state that represents p, p is true, and the information-processing role of the state, qua representation of p, depends on p’s being true. Generative grammar, as conceived by Chomsky and, I hazard, the vast majority of those linguists who work within the Chomskian paradigm, is a branch of cognitive psychology, and that conception is understood to carry with it a commitment to what is called the Computational Theory of Mind (CTM)—the theory, with which I have no quarrel, that human information processing is computation in a lingua mentis, a brain’s neural language of thought, which I’ll call Mentalese. The notion of “tacit knowledge” to which GGH appeals is commonly understood to be a notion whose proper explication presupposes CTM. I shall accept CTM as a working hypothesis of this essay and will therefore assume that we process information in a neural system of mental representation, the “language of the brain,” and that for a subpersonal state to represent p it must contain a token of a Mentalese sentence that represents p. Consider, for example, the information-processing sequence that begins with a hearer’s auditory perception of the utterance of a sentence and terminates in her knowing what the speaker meant in producing the utterance. The picture implied by GGH is that crucially implicated in that 2 For ‘subdoxastic’, see Stich (1978), and for ‘subpersonal’, see Dennett (1969). 3 processing sequence will be two intermediate states: first, a state that represents the generative grammar of the language (or part of a language) the hearer shares with the speaker, and at some point after that a state that represents the sounds produced by the speaker as tokening a sentence with syntactic structure Σ and semantic interpretation M, for some structure Σ generated by the grammar’s syntactical component and some interpretation M that the grammar’s semantic component assigns to Σ. This is another intuitively attractive picture, for an apparent platitude correlative to the one already mentioned is that one knows what a speaker means in uttering a sentence at least partly on the basis of knowing what the sentence he uttered means, and how could one compute the meaning of a sentence unless one’s processing accessed a grammar that generated its meaning? Much of this essay will be concerned with the question of what must be true of a semantic theory in order for it to be the semantic component of a generative grammar that verifies GGH. Now, there are broader and narrower ways of talking about “generative grammar,” and some of these broader or narrower ways may be fueled by questionable stances on substantive theoretical questions. To take one prominent example,3 Chomsky identifies a person’s I-language with the internally represented generative grammar that results when the parameters in the universal grammar (UG) that is innately represented in the mind/brain of every normal child are caused by the very young child’s exposure to the speech of those in her linguistic environment to take on the specific values that determine her to have out of all the possible languages she might have acquired the particular language she does acquire. But the information that can be used to fix the language-determining parameters of UG are, according to Chomsky, rather severely constrained by the nature of the language faculty (LF) and can’t be information that can be acquired only by interactions between outputs of LF and the outputs of other cognitive and motor faculties. Consequently, the properties assigned to an expression by what he would call a generative grammar won’t involve relations between the expression and external-world objects and properties, properties that would be needed to determine the expression’s extension and enable speakers to use the expression in communication, but are instead “internalist” syntactic properties—albeit syntactic properties Chomsky 3 See e.g. Chomsky (1995). 4 pointedly calls “semantic” without denying that they are also syntactic—which explain how expressions with those properties are able to interact with other cognitive faculties in ways that enable speakers to have and utilize tacit knowledge about the properties their expressions acquire as a result of their use in communicative and other behavior, properties that are essential to what we pre-theoretically might regard as essential to language understanding.